Why Have Housing Prices Gone Up?

Transcription

1 Why Have Housing Prices Gone Up? by Edward L. Glaeser, Harvard University and NBE Joseph Gyourko University of Pennsylvania and aven E. Saks Harvard University raft of ecember 28, 2004 Abstract Since 1950, housing prices have risen regularly by almost two percent per year. Between 1950 and 1970, this increase reflects rising housing quality and construction costs. Since 1970, this increase reflects the increasing difficulty of obtaining regulatory approval for building new homes. In this paper, we present a simple model of regulatory approval that suggests a number of explanations for this change including changing judicial tastes, decreasing ability to bribe regulators, rising incomes and greater tastes for amenities, and improvements in the ability of homeowners to organize and influence local decisions. Our preliminary evidence suggests that there was a significant increase in the ability of local residents to block new projects and a change of cities from urban growth machines to homeowners cooperatives. 1

2 I. he ise in Housing Prices he mean and variance of housing prices have risen across the United States since he top panel of able 1 shows that the average price across the 316 metropolitan areas of the continental United States has increased 1.7 percent annually from $59,575 in 1950 (in 2000 dollars to $138,601 in More notable is the widening variance. Since 1970, the standard deviation of real prices across metropolitan areas increased by 247 percent compared with a 72 percent increase in average prices. As Figure 1 shows, this rising variance reflects an explosion of housing values at the top end of the price distribution. he top line in the figure plots the real average price for the metropolitan area at the 90 th percentile of the house value distribution. he second line depicts price in the median metropolitan area, and the third line shows the mean house value for the area at the 10 th percentile of the distribution. In 1970, the average house price in the metropolitan area at the 90 th percentile was 35 percent more expensive than that of the median metropolitan area. In 1990, the 90 th percentile area s price was more than twice as expensive as for the median metropolitan area. here has been little change in the gap between the median metropolitan area and low-cost areas. oo often, analysts attempt to understand housing prices only by attending to demand-side factors such as interest rates or per capita income, while ignoring the supply-side of the market. ising prices require not only rising demand, but also limits on supply. he supply of housing includes three elements: land, a physical structure, and government approval to put the structure on the land. hus, rising prices must reflect rising physical costs of construction, increasing land prices or regulatory barriers to new construction. 2

3 he bottom panel of able 1 reports the real value of construction costs per square foot for a modest-quality, single-family home in a sample of 177 markets tracked by the.s. Means ompany, a data provider to the home building industry. Average house prices and construction costs rose together between 1950 and 1970, but since 1970, the cost of putting up the physical structure has declined slightly while housing prices have continued to rise. Even in booming markets, construction cost increases have been modest. Between 1970 and 2000, real construction costs in San Francisco and Boston rose by 4.6 percent and 6.6 percent, respectively. Over the same 30 years, real mean house prices rose by 270 percent in the San Francisco primary metropolitan statistical area (PMSA and 127 percent in the Boston metropolitan area. ising structure costs still could explain the post-1970 growth in housing prices if structural size and quality were increasing rapidly. o assess the importance of housing quality, we can compare the overall rise in housing prices with the rise in prices measured by repeat-sales indices that hold housing structure constant. In the United States as a whole, the real median value of owner-occupied housing rose by 1.20 percent per year from 1980 to Over this same period, real appreciation of the repeat-sales index published by the Office of Federal Enterprise Housing Oversight (OFHEO was 0.93 percent per year, which suggests that changes in the quality of housing account for no more than one quarter of the average increase in housing values. he same methodology in high-price areas shows that quality growth is even less important in those places. 1 1 For example, real median prices rose by 3.14 percent per year between in the San Francisco metropolitan area according to U.S. ensus data, while OFHEO reports a 3.45 percent real appreciation rate for its constant quality series of repeat sales. he analogous numbers for the New York metropolitan area are 3.64 percent and 3.71 percent, respectively. Because changes in structure quality clearly do not account for higher prices in the most expensive coastal markets, we use ensus data because it is available over a much longer time period. 3

4 A summary measure of the importance of physical structure is the ratio of the average house price (P to the estimated physical construction cost ( for 102 metropolitan areas in each ensus year from , 3 If the housing supply market is competitive, then the difference between the value of this P/ ratio and one tells us how much of the value of housing cannot be accounted for by the physical cost of supplying the unit. 4 his remainder reflects either the cost of land or the costs of obtaining regulatory approval. he top two rows of able 2 report the sample means and standard deviations of the distribution of P/ in each of the last six ensus years, while the bottom two rows report the implied share of land and regulatory approval (i.e., 1 /P. Even in the most expensive metropolitan areas, structure appears to have represented almost all of the cost of housing in 1970 and earlier years. 5 As the bottom two rows of able 2 indicate, the physical cost of constructing the house represented about 90 percent of the value of the home in the metropolitan area in the 90 th percentile of the P/ distribution in In the San Francisco PMSA, which had the lowest share 2 Because construction costs are reported per square foot, housing values are divided by an estimate of the median size of single-family homes in each metropolitan area. Specifically, the American Housing Survey (AHS is used to calculate median unit size for single-family homes in We then assume that unit size grew by 2.75% per decade between 1960 and 2000, which roughly corresponds to national changes in unit sizes in the AHS and information on the size of new single-family homes. In other work using data dating back only to 1980, we make additional adjustments to the numerator and denominator of the P/ ratio to help control for potential bias in self-reporting of house values and for aging of the housing stock (see Glaeser and Gyourko (2005. Given the longer time span examined here, we are unable to make those same adjustments. Our previous research shows that not controlling for the depreciation on existing homes is more important empirically than not controlling for the upward bias in self-reported house values. Hence, the P/ ratios used here are likely to be biased downward. Experimentation confined to data from more recent decades finds that the cross sectional distribution of P/ ratios across cities is not sensitive to these adjustments (or lack thereof. 3 hese 102 areas include the PMSA components of MSAs. Grouping the relevant PMSAs together, there are 68 distinct MSAs, MSAs and NEMAs. 4 Glaeser, Gyourko and Saks (2005 present extensive evidence suggesting the highly competitive nature of the housing market. 5 here are several explanations for why we observe so many metropolitan areas with average housing price below construction costs. As mentioned above, these estimates are not adjusted for the depreciation of existing homes. Another factor that would bias our estimates downward is if housing units were smaller in the past than we assume. In general, we have chosen assumptions to err on the side of yielding a conservative (lower P/ ratio. 4

5 of total house value accounted for by physical construction costs, only 24 percent was not due to structure value. It is only since 1980, and only in a relatively few metropolitan areas, that there has been a widening gap between price and construction cost. Almost all of the markets in which housing prices became substantially higher than physical production costs during the 1970s were part of the three big coastal metropolises in alifornia the Los Angeles-centered MSA, the San Francisco-centered MSA, and the San iego MSA. A decade later, gaps between prices and construction costs on the West oast had grown and had spread to interior markets in alifornia. For example, structure represented only 53 percent of average house value in Sacramento in High prices relative to construction costs had also spread to other West oast markets such as Seattle. he non-structure component of house value also exceeded 40 percent across a swath of the east coast roughly approximated by Amtrak s Northeast orridor. By the year 2000, there were 27 metropolitan areas in which structure could account for no more than 60 percent of total house value. hese locations include virtually all of the coastal areas already discussed. Moreover, the 1990s witnessed the spread of high housing prices relative to physical production costs to interior markets such as Ann Arbor, MI, Austin-San Marcos, X, enver, O, Nashville, N, and aleigh-urham-hapel Hill, N. In the highest priced housing markets in the nation (the PMSAs within the San Francisco MSA, structure is estimated to represent no more than 30 percent of house value. Figure 2 plots the P/ ratios in 2000 versus 1970 for all 102 metropolitan areas, and starkly illustrates the dramatic rise in the gap between price and construction cost in the highest price markets over the past three decades. 5

6 he key to understanding the rise in housing prices relative to construction costs is that new construction has declined sharply in high price locations. Figure 3 documents the drop in new construction intensity in three increasingly expensive metropolitan areas: New York, San Francisco and Los Angeles. New construction is measured at decadal frequencies as the share of housing units built since the last ensus divided by the number of units in the area in the preceding ensus. he values reported for 1960 indicate that Los Angeles increased its stock by nearly 60 percent during the 1950s, while the number of housing units in San Francisco expanded by more than 30 percent. Even New York increased the size of its housing stock by more than 20 percent that decade. By the 1990s, the housing stock in all three metropolitan areas increased by well under 10 percent over the decade. Although these three large, high housing price areas are outliers, able 3 shows that there has been a significant reduction in the rate of new construction nationally. In the 1950s, the median rate of new construction was 40 percent in our sample of 102 metropolitan areas. Four decades later it had fallen to 14 percent, which was still double the rates seen in San Francisco, New York, and Los Angeles. As late as the 1970s, there was a robust relationship between new construction and the ratio of price to construction costs. Figure 4 shows that in places where prices were high relative to construction costs in 1970, there was more new construction over the ensuing decade (with the notable exception of the San Francisco PMSA. Figure 5 repeats the exercise for the 1990s, documenting that this basic relationship has been reversed. It is no longer the case that high prices relative to construction costs generally lead to a surge in new construction. 6

7 able 4 shows the declines in residential construction more rigorously. For our panel of 102 metropolitan areas, we begin by regressing the ratio of new construction to the initial housing stock on year dummy variables. hese coefficients, which are reported in the first column, document the significant decline in the intensity of new construction between 1960 and today. he second regression includes controls for density, income and the ratio of price-to-construction cost at the beginning of the decade. he controls for price and density are both statistically significant, but they explain little of the change over time. hese three controls together reduce the coefficient on the year 2000 by.05, or about 14 percent. While such a regression is only suggestive, the results indicate that rising density levels can only explain a small amount of the decline in new construction over time. he third regression adds interactions of the P/ variable with the year dummies. Just as the figures suggested, there was a powerful relationship between price and new construction during the earlier time periods that vanished by the 1990s (e.g., the total effect of P/ in 1990, which is the sum of coefficient on the level of P/ and its interaction with 1990, is very close to zero. hese results strongly suggest that restrictions on new supply have become increasingly important in preventing suppliers from responding to high prices by building additional units. But are these limits on new construction the result of a dwindling supply of land or other barriers to new construction? able 4 already suggested that this change cannot be explained by density alone. Further evidence that rising density is not strong enough to explain large declines in construction can be found by examining individual permit-issuing places within the San Francisco metropolitan area. Figure 6 depicts a robust negative relationship between initial period density and new construction in the 7

8 1990s. However, it also shows many low-density, high cost areas with very little new construction. 6 A final check on the density hypothesis is to return to the share of housing prices that is not related to physical construction costs. Beyond physical structure, the cost of supplying a house includes both the cost of the land and the cost of the right to build. If the costs associated with the right to build were small, then the non-structure value of the property would include only the cost of the land. A completely free market for land would lead land to be worth the same amount on both the intensive and extensive margins. Stated differently, a quarter acre would be valued the same if it sits under one house or if it extends the lot of another house. Using this insight, Glaeser and Gyourko (2003 and Glaeser, Gyourko and Saks (2005 use hedonic price estimation to estimate the price of land when it extends the lot of an existing house. We find that such land is not all that valuable; generally a quarter acre is worth about ten times more if it sits under a house than if it extends the lot of another house. he fact that land is worth much more when it is bundled together with the right to build provides further evidence that the right to build is worth a great deal. 7 In sum, the evidence points toward a man-made scarcity of housing in the sense that the housing supply has been constrained by government regulation as opposed to 6 A very similar relationship holds if one looks at the same relationship at the census tract level. hat is, more dense tracts have lower permitting intensity, but there are plenty of less dense tracts in the San Francisco area with equally low permitting activity. hat graph is available upon request. More generally, Glaeser and Gyourko (2003 show that there is little relationship between density and high housing costs across U.S. metropolitan areas. 7 Additional evidence on the role of zoning is given by the relationship across metropolitan areas between zoning and reduced construction levels and higher price increases. Saks (2004 uses an index of subjective assessments of zoning regulations and finds a positive relationship with housing prices and a negative relationship with residential construction. 8

9 fundamental geographic limitations. 8 he growing dispersion of housing prices relative to construction costs suggests that these regulations have spread into a larger number of local markets over time. Moreover, they appear to have become particularly severe in the past 2-3 decades. o explore the source of these changes, the next section develops a model of the incentives faced by local governments to impose restrictions on residential development. II. he Economics of Zoning and Permitting Our model is one of a local zoning authority that decides whether to approve or reject residential development. here are two locations: the zoning authority s town and a reservation locale. here are N total consumers, of which live in the town. he remainder of the population lives in the reservation locale, and there are no constraints preventing people from moving there. otal utility in this outlying area is a decreasing function of the number of people living there, U ( N. In the town, the flow of utility equals U(+a-housing costs, where U( is decreasing in the amount of development in the city, and a is an individual-specific desire to live in the locale. he distribution of a is described by a cumulative density F(a and density f(a. he cost of construction in the town equals K, which can be thought of as capturing both physical costs of construction and the opportunity cost of land taken away from agricultural uses. We normalize the cost of construction in the reservation locale to be equal to zero, so that K reflects the additional cost of building a housing unit in town. enoting the interest rate as r, the annual cost of housing construction is rk. 8 Even in areas like Manhattan where water restricts the ability to build outward, the option to build upward remains. he choice to limit building heights constrains the number of new housing units just as other zoning regulations may limit the number of single-family houses. 9

10 As in any spatial equilibrium, there will be a marginal consumer with a taste for the town equal to â who is indifferent between living in the town or the reservation locale. Every consumer with a value of a greater than â will live in the town and the remaining consumers will live in the reservation locale. he marginal consumer must satisfy N( 1 F( aˆ 1 = ; we use the notation a ˆ( F ( 1 / N =. he initial population of the town is split into homeowners and renters. We assume that a fraction h of these units are allocated to homeowners and the remainder to renters. All individuals are assumed to live in the community for exactly L time periods. After that time, individuals are replaced by identical consumers so that the total size of the population remains unchanged. Individuals maximize V rt e u( t t= 0 dt rv + e Asset V where u(t is the flow of utility at time t, r is the interest rate and Asset V is the value of any asset as of time V. enters pay the market clearing rent, which is equal to the same annual cost as the interest payments on a house. Both rents and housing values will decline with new development. If the town starts with housing units, then houses in town will be worth U ( U ( N + aˆ( r. Given these assumptions, there is a unique amount of development that will maximize the average discounted lifetime utility of all current residents of the town. In order to achieve this social optimum, it is necessary to allow for the possibility of sidepayments between developers (who gain from additional residential construction and homeowners (who lose from additional development through lower future housing values. It is straightforward to show that a higher fraction of homeowners will lead to less development. Moreover, because a shorter lifespan makes the resale value of homes 10

11 more salient to homeowners, shorter life spans also curtail the optimal amount of development. here are two reasons why the level of development that maximizes the welfare of current residents will not be socially optimal. First, higher population density has a negative impact on the utility of future residents of the town and of the reservation locale that current residents will not internalize. Second, current homeowners have an incentive to increase the value of their homes, and do not internalize the impact that higher housing prices have on non-homeowners who would like to live in the town. We now consider the decision faced by the town s zoning authority who decides whether a new development project of size will proceed. We simplify the analysis by assuming that utility in the reservation locale is fixed at U. Furthermore, we will ignore the incentive of renters to lobby for more housing to be built. Essentially, this assumption implies that renters are not organized enough to support the construction of new housing. he zoning authority will receive net benefits of α + g + g ( + ε from rejecting the project. he parameter α ( captures the innate distaste of the authority for development. and reflect the cash spent by developers and town residents to influence the authority s decision, and g is a concave function reflecting the influence that cash will have on the decision-making of the authority. Similarly, and reflect the time spent by the developer and residents, respectively, on influencing the authority, with a concave function g representing the influence of time on the authority. We assume that both g(. functions are symmetric around zero. Finally, ε is a uniformly distributed mean-zero idiosyncratic term with 11

12 density 1. We will assume that parameter values are such that there is always some positive probability that the project will be both accepted and rejected. project equals developer as Under these assumptions, the probability that the authority will authorize the 5 + g ( + g ( α. We denote the cost of time to the. W and the cost to residents as W. he developer therefore chooses the amount of time and cash spent to influence the zoning board to maximize: U ( + U + aˆ( +. α. r ( 5 + g ( + g ( K W From the perspective of each current homeowner, the development project will create a net loss equal to ( aˆ( + aˆ( rl U ( + U ( + e r. his expression reflects both the negative externality associated with increased population density in town and the decline in housing values. Individuals face costs of influencing the authority equal to W +. Because a continuous distribution of residents implies that each individual person has a negligible impact on the zoning decision, we assume the existence of a community organization that organizes town residents. his organization includes a proportion λ of homeowners and maximizes the aggregate utility of its members. We assume that W > W so that the opportunity cost of time is higher to the developer than to the homeowners. Although this assumption seems plausible, it implies that landlords cannot employ renters to lobby the zoning board at the same time cost faced by homeowners. 9 9 Historically, it has been rare to see renters fight zoning restrictions. Perhaps this is due to some agency problem that prevents developers from following this strategy. Stronger homeowner participation might be because homeowners simply enjoy the social activity of protesting new developments. 12

13 the appendix: aken together, this model implies the following two propositions (proofs are in Proposition 1: If both the landlord and the homeowners association undertake some lobbying effort, then the landlord will use only cash and the homeowners will use only time. Such specialization of effort seems consistent with much anecdotal evidence on local battles between developers and community groups (Warner and Molotch (2000. his result implies: Proposition 2: If both actors engage in a positive amount of lobbying then: (1 the probability that the project will be approved declines with α, (2 the probability that the project will be approved is decreasing with h and λ, (3 if g ( x = γ g~ ( x and g ( x = γ g~ ( x then the probability the project will be approved declines with γ and rises with γ, and (4 if U ( + = U ( u, then the probability the project will be approved falls with u. Proposition 2 sets forth a number of comparative statics that can potentially explain the change in the zoning environment in the United States. he first, and perhaps most popular explanation to date, is that an increase in zoning reflects changes in the preferences of judges and other political decision-makers. While there is evidence from legal and economics scholarship that judges and local government officials have become increasingly sympathetic to community and environmental concerns (see the discussion below, it is unsatisfying to explain a large-scale shift of this nature simply by appealing to changing preferences. 13

14 he second comparative static suggests that the explanation lies in the rise of homeownership and the success of community organization. Increases in both the share of homeowners and the political organization of homeowners groups should lead to less development. In the past 40 years, the fraction of homeownership has risen from about 59 to 68% (Federal eserve Board, 1964 and Aizcorbe et. al., Moreover, political participation of homeowners groups has been rising (Nelson, 2004, Freund, Not only should this trend restrict residential development, but Altshuler and Luberoff (2002 suggest that these groups have been increasingly able to restrict large-scale nonresidential development projects as well. he third part of Proposition 2 points to the changes in the relative effectiveness of using cash versus time to influence political decision-makers. γ can be interpreted as the efficacy of bribes, and it is quite plausible that this parameter has declined over time. In contrast, it is likely that the efficacy of spending time to influence decision-makers has increased. ising education levels and learning from other political battles (e.g., the civil rights movement may have made community members more savvy about using courts and the press. he final comparative static concerns the taste for density. If rising incomes have caused people to place a higher value on living in a low density community, then we should expect to see less development. Other factors, such as crime and improvements in transportation, may also have increased the desirability of low-density living. III. Evaluating the Explanations for a More estrictive Zoning Environment In this section, we review the possible reasons why it has become more difficult to build new homes since

15 Judicial astes In 1977, obert Ellickson noted that suburban governments are becoming ever more adventuresome in their efforts to control housing development. (p Ellickson does not explain this change, but points to judicial decisions such as Nectow v. ity of ambridge which have made it difficult for landowners to stop municipalities from restricting new construction on their land. Fischel (2004 points to the ideology of judges: ourts, whose judges share the same environmental attitudes as middle class homeowners (just as 1920s judges shared the ideology of hearth and home, were more sympathetic to claims that the local decision had failed to account for environmental impacts than they had been to seemingly selfish claims that neighbors home values were at risk. (pp Other cases such as Mt. Laurel that demanded low income housing have simultaneously allowed growth controls: "once a community has satisfied its fair share obligation [a fraction of the region s low-income housing], the Mount Laurel octrine will not restrict other measures, including large-lot and open area zoning, that would maintain its beauty and communal character"(mount Laurel II, 456 A.2d at 421 cited in Fischel (2004, p here can be little doubt that court decisions have become friendlier to antidevelopment sentiment. While courts clearly are important, ultimately it is unsatisfying to attribute the change in the zoning environment to changing attitudes of judicial decision-makers. hese attitudes are not exogenous, but reflect other trends in American society. If changes in the tastes of judges and policy makers reflect societal trends like the environmental movement, then these changes should be viewed as an improved effectiveness of certain groups in shaping policy. In the language of the model, this 15

16 should be viewed as an increase in or γ, not an exogenous change in α. Empirically, we cannot reject the hypothesis that judicial tastes changed, but on theoretical grounds this explanation is so unsatisfying that we will turn elsewhere. he Impact of esidents Groups While the influence of developers may or may not have declined, many observers have noted a sizable increase in the organization and political impact of local residents. Altshuler and Luberoff (2002 examine the history of large scale government projects ( Mega Projects and describe changes that began in the 1960s, when citizens became better able to challenge large scale projects that would impact their neighborhood. One early and striking example was Jane Jacobs leadership of the Greenwich Village movement that stopped obert Moses West Side highway project in New York. hrough increasingly sophisticated use of the media, local groups learned how turn mega-projects into public relations disasters. here is abundant evidence of the impact of homeowners and neighborhood groups, but there is less understanding of where this impact comes from. One hypothesis is that homeowners have become better organized (an increase in λ. Some analysts have suggested that the organization skills of environmental groups were learned from the organizational successes of the civil rights movement and the anti-war protests. Either through imitation of these earlier groups or because of rising education and media savvy, local residents appear to have become better at using the media and the courts. hus, the typical residential activist of 2004 seems more skilled than its counterpart from

17 he Ability to Use ash to Influence Local ecision-makers A third possible hypothesis is that developers ability to use cash to influence local decision-makers has fallen over time. his influence historically has come both from legal payments, in the form of campaign donations or legal cash transfers (i.e. a developer employing a politician for legal work, or illegal cash payments or bribes. Zoning environments may have become more restrictive if developers in the 1960s were more easily able to bribe local politicians than they can today. In other words, the urban growth machine described by Molotch (1976 has weakened as it has become harder for developers to transfer cash to politicians. here is some evidence suggesting a decline in corruption over time within the United States. Glaeser and Goldin (2004 use newspaper records to show a decline in the share of articles alleging corruption between the late 19 th century and the mid-20 th century. However, their coverage does not show a significant change between 1960s and the 1990s the period of the permitting slowdown. Anecdotes about corruption in development abound, and it may be true that such anecdotes were more common in the 1960s than today. While this hypothesis remains plausible, there is precious little evidence either supporting or refuting it. Even if it were possible to show such a change, it would be desirable to go further and try to understand why this change occurred. One plausible explanation is that improvements in the news media have caused more attention to be paid to corrupt deals. A second explanation is that the political influence of local party machines has declined. hese machines facilitated the flow of funds from developers (or anyone else and ensured that legal repercussions from local justice would be modest. he decline of local 17

18 machines might also have played a role in reducing the influence that developers were able to have on local governments. he Value of Amenities Another natural explanation for the rise in restrictions on new construction is that rising income levels have increased the willingness to pay for high amenity neighborhoods, and in particular, for low density neighborhoods (assuming low density is a normal good, of course. his hypothesis corresponds to an increase in the parameter u, which the model predicts should lead to a decrease in permitting as the incentive of homeowners to spend time to block new construction rises. o evaluate the importance of rising incomes in explaining the decline in permitting, we regress the share of new housing units in 1960 on the ratio of housing prices to construction costs, density, and the logarithm of income in he coefficient on income is -.14 (with a standard error of.23, indicating that richer communities were less likely to build new housing units. However, the magnitude of this coefficient is not large enough to explain the general decline in permitting over time. Our data suggest that permitting has declined by 37 percentage points between 1960 and today. In those same areas, real median incomes have risen by.77 log points. Using the estimated coefficient on income, together these values suggest that rising incomes can explain only about 29 percent of the fall in residential construction. ising American incomes are important, but they are only a small part of the change in the permitting environment. Another way to think about the effect of income is to consider the zoning environment of very rich places in If the income hypothesis is correct, then 18

19 permitting in these places should have been as restrictive in 1960 as the entire metropolitan areas of Boston or New York in more recent years. However, places like New ochelle, NY, San Mateo, A and West Orange, NJ, each allowed at least 10 times as much development in the 1950s as metropolitan areas with comparable incomes today. Again, this analysis suggests that the complete story goes well beyond the explanation that homeowners became richer. hanges in the Housing Market A final hypothesis is that the impact of new construction on housing prices has changed over time. In the 1950s, housing costs were low, lower incomes made people less concerned about environmental amenities, and an absence of construction in previous decades may have meant that the quality of new housing was significantly higher than older units. For these reasons, new construction may not have led to major reductions in housing prices for existing units and as such, homeowners had much weaker incentives to fight new construction. In 2004, however, homeowners appear to believe that new construction will significantly reduce housing prices. ertainly, the evidence in this paper linking rising housing prices to reductions in construction suggests that they are right. As in the case of the previous theories, we have little evidence on the relevance of this theory and we look to further research to examine this hypothesis more thoroughly. IV. onclusion Housing is one of the most important elements in household portfolios and budgets. Over the past 30 years, the dispersion in prices across American markets has increased substantially. In much of the country, new housing units still are abundant and 19

20 housing prices remain low. In contrast, new construction has plummeted and housing prices have soared in a small, but increasing number of places. hese changes do not appear to be the result of a declining availability of land, but rather are the result of a changing regulatory regime that has made large-scale development increasingly difficult in expensive regions of the country. hanges in housing supply regulations may be the most important transformation that has happened in the American housing market since the development of the automobile, but they are both under-studied and under-debated. he positive research agenda going forward should be to understand why these changes have occurred and what relationship exists with other major trends in the American society. he normative policy agenda should be to better understand the costs and benefits of limits on new construction. he costs appear to include higher prices and a misallocation of labor, while the benefits include internalization of construction-related externalities. Given the implications of this regulatory shift, the economics profession could make a major contribution by analyzing the welfare effects of regulation on the rise in housing prices. 20

21 eferences Aizcorbe, Ana, Arthur B. Kennickell and Kevin B. Moore ecent hanges in U.S. Family Finances: Evidence from the 1998 and 2001 Survey of onsumer Finances. Federal eserve Bulletin (January. Altshuler, Alan and avid Luberoff Mega-Projects. Washington, : he Brookings Institution. Ellickson, obert Suburban Growth ontrols: An Economic and Legal Analysis. Yale Law Journal 86 (3: Federal eserve Board of Governors Survey of Financial haracteristics of onsumers. eport (March. Fischel, William An Economic History of Zoning and a ure for Its Exclusionary Effects. Urban Studies 41(2: Freund, Eric Land evelopment Management: evolution and Evolution in Municipal Year Book. Washington: International ity Management Association. Glaeser, Edward and Goldin, laudia orruption and eform: An Introduction. National Bureau of Economic esearch Working Paper Glaeser, Edward and Joseph Gyourko he Impact of Zoning on Housing Affordability. Economic Policy eview 9(2: Glaeser, Edward, Joseph Gyourko and aven E. Saks Why is Manhattan So Expensive? egulation and the ise in House Prices. Journal of Law and Economics. Gyourko, Mayer, and Sinai Superstar ities. Zell/Lurie eal Estate enter at Wharton Working Paper, July 2004, University of Pennsylvania. Molotch, Harvey he ity as a Growth Machine. American Journal of Sociology 82(2: Nelson, obert H he ise of Private Neighborhood Associations: A onstitutional evolution in Local Government. prepared for a conference on he Property ax, Land Use and Land-Use egulation, sponsored by the Lincoln Institute of Land Policy, January 13-15, S. Means ompany. 2000a. esidential ost ata. Kingston, MA :.S. Means o. 2000b. Square Foot osts. Kingston, MA :.S. Means o. 21

22 Saks, aven E Job reation and Housing onstruction: onstraints on Employment Growth in Metropolitan Areas. Working Paper. Warner, Kee and Molotch, Harvey Building ules: How Local ontrols Shape ommunity Environments and Economies. Boulder, O: Westview. 22

23 Appendix: Proofs of Propositions Proof of Proposition 1: We use the notation that Φ U = rl ( U ( + + e ( aˆ ( aˆ( + r and Φ U ( + U + aˆ( + = K r for the losses and gains from development for the homeowners and the developer respectively. For the developer, the marginal product of spending an additional dollar is g ( Φ 1 and the marginal product of spending more time equals g ( Φ W. he second derivative of the developer s utility with respect to spending is g ( Φ and the second derivative of the developer s utility with respect to time is g ( Φ. For the homeowners, the marginal impact of more spending is λ hφ g ( 1 and the marginal impact of more time is λ hφ g ( W. he second derivative with respect to spending is λ hφ g and the second derivative with respect to time is λ hφ ( g (. Obviously, if the second order conditions hold for the homeowners they cannot hold for the developer and vice-versa. Because the problem is inevitably concave for one of the actors and convex for the other, it cannot be an optimum in which for both actors use both technologies. We have assumed that each actor uses at least one technology, so it remains to be shown which technology will be used by each actor. For the homeowners to use cash, it must be the case that λ hφ g ( W < 0 = λhφ g ( 1 which would imply that g ( / g ( < W. For the developers to use time, 23

24 g ( Φ 1 < 0 = g ( Φ W, but this would imply that g ( / g ( > W which is a contradiction. On the other hand, if the homeowners use time and the developers use cash, this must imply that λ hφ g ( W = 0 > λhφ g ( 1 and g ( Φ 1 = 0 > g ( Φ W. Proof of Proposition 2: Given the assumptions that have been made, the probability that the project is authorized g ( Φ = g ( + g ( α, where λ h Φ g ( = W and. (1 Given that the value of α does not impact the type of lobbying employed by either interest group, the probability of the project being approved will decline with α. (2 ifferentiation of λ h Φ g ( = W and second order conditions implies that is rising with both λ and h. As rises, the probability that the project will be approved falls. (3 If g ( x = γ g~ ( x then the probability of approval equals 5 + γ ~ g ( + g ( α and the derivative of this with respect to γ is. g ~ ( + γ g~ ( γ, where using the implicit function theorem, γ = g~ ( γ g~ ( which is positive since second order conditions have been assumed to hold, so the overall impact of γ is positive. If g ~ ( x = γ g ( x then the derivative of the probability of approval with respect to γ equals g ~ ( γ g~ ( γ, 24

25 25 where 0 ( ~ < g and ( ~ ( ~ g g = γ γ which is positive since second order condition have been assumed, so the overall impact is negative. (4 If = + u U U ( (, then the parameter u only enters through the terms ( ( ( r a a e U U rl ˆ( ( ˆ = Φ and = Φ K r a U U ˆ( (. ifferentiation then yields the result that 0 ( ( > Φ = g r g u and 0 ( ( 2 < Φ = g g u, so lobbying by homeowners increases and lobbying by developers decreases. hus, the probability of approval must fall.

26 able 1: eal House Prices Over ime 316 Metropolitan Areas; $ Mean $59,575 $73,741 $80,556 $109,570 $120,929 $138,601 % hange in Mean Over ecade % 9.2% 36.0% 10.4% 14.6% eal Single-Family onstruction osts Over ime One-Story, Modest Quality Home $2000/ft Mean $49.70 $58.50 $63.60 $65.40 $63.30 $61.60 % hange in -- Mean Over 17.7% 8.7% 2.8% -3.2% -2.7% ecade Notes: 1. he source for the house price data is Gyourko, Mayer, and Sinai (2004. Prices are for single-family homes, with metropolitan area values being aggregated across their county components based on 1999 definitions from the Office of Management and Budget. Prices are reported for individual primary metropolitan statistical area (PMSA components of consolidated metropolitan statistical areas (MSAs. 2. onstruction costs are from the.s. Means ompany, a consultant and data provider to the home building industry. he underlying data include material costs, labor costs, and equipment costs for a lower-quality, one story house without a basement that still meets building code requirements in each market. See the following two publications by the. S. Means ompany for greater detail on the underlying cost data: esidential ost ata, 19 th annual edition, (2000 and Square Foot osts, 21 st annual edition (2000, both published by the.s. Means ompany. 26

27 able 2: Price-to-onstruction ost atios (P/ Over ime 102 Metropolitan Areas Mean Standard eviation th Percentile Maximum Implied Land Share (~1 /P 90 th Percentile Maximum Notes: Mean house prices are constructed for each metropolitan area using county-level data from the relevant decennial census. onstruction cost data are from the.s. Means ompany as described above. Various adjustments to the price and cost data as described in the text. 27

28 able 3: New Housing Units as a Fraction of Initial Housing Stock 25 th percentile Median 75 th percentile Note. New units in 1960, 1970 and 1980 are the number of housing units built since the previous census. New units in 1990 and 2000 are the total number of building permits. Sample is limited to a balanced panel of 102 metropolitan areas for which it is possible to calculate housing prices relative to construction costs. 28

29 able 4: eclines in esidential onstruction ontrolling for ensity onstant ( ( ( (.07 Year ( ( ( ( ( ( ( ( ( ( ( (.08 Year* P/ t ( ( ( (.28 Log density t ( (.02 P/ t ( (.30 Log income t ( (.14 Adjusted Note. ependent variable is the number of new units in each decade relative to the initial housing stock. New units in 1960, 1970 and 1980 are the number of housing units built since the previous census. New units in 1990 and 2000 are the total number of building permits. Sample is limited to a balanced panel of 102 metropolitan areas for which it is possible to calculate housing prices relative to construction costs. ensity is the number of housing units per square mile. Income is a weighted average of median family income by county, where the weights are the fraction of families in the metropolitan area. Each right-hand side variable is expressed relative to its sample average. Standard errors are clustered by metropolitan area. 29

30 2000 $ Figure 1: hanges Across the House Price istribution Average Values, 316 MSAs year 10th 50th 90th 30

31 Figure 2: House Prices elative to onstruction osts in 1970 and San Jose San Fran Housing Price/onstruction osts, Scranton Allentow Santa r Oakland, NashvillNew Have Fort Lau Yolo, A Washingt aleigh- Austin-S Sacramen Ann Arbo Bergen-P Boston-W Atlanta, Miami, Newark, FNew Nassau-S York Greeley, Albuquer Salt Lak harlest Middlese Norfolk- olorado Baltimor Greensbo iversid Phoenix- Monmouth Springfi hicago, olumbia Bremerto allas, olumbus Memphis, Jacksonv Hamilton Las Vega New Orle Olympia, Orlando, ampa-st acoma, Wilmingt hattano incinna Omaha, etroit, Baton Milwauke olittle Grand Knoxvill aindianap Hagersto anton-m Kenosha, Shrevepo N Minneapo ulsa, Akron, O Fort Birmingh Wor Providen Wichita, Outchess Gary, IN levelan El Paso, Bakersfi Brazoria Houston, Peoria-P acine, Fresno, Philadel Galvesto Newburgh St. Oklahoma Kansas Albany-S Kankakee Loui San oledo, Anto Youngsto ocheste Vineland Pittsbur Syracuse Ventura, San ieg Seattle- enver, Orange Los Ange Housing Price/onstruction osts,

32 Figure 3: eclining onstruction Intensity in Select High House Value Markets San Francisco, Los Angeles, and New York PMSAs, San Francisco PMSA Los Angeles PMSA New York PMSA New Housing Units elative to Initial Stock year 32

33 Figure 4: esidential onstruction and P/ in the 1970s Metropolitan Areas with P/ > 1 in 1970 New Units elative to Housing Stock in Oakland, Norfolk- New Orle New Have Nassau-S Bergen-P New York Washingt Los Ange Orange San Jose Fort Lau San Fran P/ atio in

34 Figure 5: esidential onstruction and P/ in the 1990s Metropolitan Areas with P/ > 1 in 1990 New Units elative to Housing Stock in Las Vega Nashvill Jacksonv Ann allas, olumbia Arbo Memphis, enver, Albuquer Greensbo Fort Seattle- Lau Bakersfi harlest Yolo, Sacramen A Washingt El Paso, Wilmingt ampa-st Norfolk- iversid Hagersto Little Newburgh Miami, Baltimor F Monmouth Middlese hicago, San ieg ocheste Philadel Albany-S New Orle utchess Boston-W Shrevepo Providen Springfi New Have Nassau-S Newark, New YorkBergen-P aleigh- Orlando, Atlanta, Phoenix- Austin-S Orange Ventura, Oakland, Santa r Los Ange San Jose San Fran P/ atio in

35 Figure 6: he Intensity of New Single Family onstruction and ensity in the San Francisco PMSA (Permit-Issuing Places, Half Moo single-family permits / SF units in Novato olma to Portola ibu ron Foster Woodside orte Ma e dwoo d Hi llsbor San afa South Sa At herto n Bri sban e Menlo Pa San arl Pa cif Mill ica ValFairfax Be lmont Larkspur Sausalit oss tow San Anse Be lveder San Burlinga Mate San Millbrae Brun aly it San Fran housing units per square mile in